Edge and Cloud Computing: Where Should Factory Data Be Processed?

Edge Computing: Processing Next to the Machine

Edge computing means processing data at the point where it is generated, right next to the machine, instead of sending it to a remote server or cloud.

Why Process at the Edge?

In a smart factory, thousands of sensors generate data every second. Sending all this data to the cloud faces challenges:

• Latency: The cloud may take 50-200 milliseconds to respond, while some operations need less than 10 milliseconds
• Bandwidth: Data volume may exceed available connection capacity
• Cost: Transferring terabytes daily to the cloud is very expensive
• Reliability: Internet outages must not stop production

What Do Edge Devices Do?

• Data Filtering: Sending only significant values instead of every raw reading
• Aggregation: Computing averages and statistics locally
• Anomaly Detection: Identifying abnormal values immediately
• Local Buffering: Saving data when connectivity is lost
• Local Rule Execution: Triggering alerts or stopping a machine without waiting for the cloud

Examples of Industrial Edge Devices

• Industrial PC: Siemens SIMATIC IPC, Beckhoff C6030
• Smart Gateways: MOXA, Advantech, Siemens IOT2050
• Economical Solutions: Raspberry Pi 4 with industrial protection
• Edge Servers: Dell Edge Gateway, HPE Edgeline

Cloud Computing: Unlimited Power

Cloud computing provides virtually unlimited computing and storage resources over the internet, billed on a pay-per-use basis.

Cloud Advantages for Manufacturing

• Scalability: Add resources instantly when needed
• Advanced Analytics: Run AI models requiring massive computing power
• Long-Term Storage: Store years of historical data at low cost
• Global Access: Monitor multiple factories from anywhere
• Continuous Updates: Cloud services evolve without local hardware updates

Industrial Cloud Services

• AWS IoT: IoT Core, SageMaker for machine learning, S3 for storage
• Azure IoT: IoT Hub, Digital Twins, Machine Learning
• Google Cloud IoT: BigQuery for analytics, Vertex AI
• Siemens MindSphere: Specialized industrial IoT platform

Common Concerns About the Cloud

When to Process Locally vs Send to the Cloud

Choosing where to process depends on the nature of the task. Here is a practical decision framework:

Process at the Edge When:

• Latency is critical: Safety systems, real-time control
• Data volume is massive: Inspection cameras sending gigabytes per minute
• Connectivity is unreliable: Remote locations or weak infrastructure
• Privacy matters: Sensitive data that must not leave the factory
• The decision is simple: Comparing a value against a threshold

Send to the Cloud When:

• Analysis is complex: Machine learning models requiring powerful GPUs
• Data is historical: Comparing performance across months or years
• Reports are managerial: Dashboards accessed by managers from different locations
• Integration is broad: Connecting data from multiple factories or partners
• Storage is long-term: Archiving quality and compliance data

Hybrid Architecture: Best of Both Worlds

A hybrid architecture combines edge computing and cloud computing to achieve the optimal balance.

How Does Hybrid Architecture Work?

Sensors → Edge Devices → [Immediate local decisions]
                       → [Summarized data] → Cloud → [Advanced analytics]

Design Principles

1. Immediate Processing at the Edge: Everything requiring fast response is processed locally
2. Summarize Before Sending: Send averages and statistics instead of every raw reading
3. Smart Local Buffering: Store data locally and transmit when connectivity is available
4. Bidirectional Synchronization: The cloud sends updated models back to the edge

Data Flow Model

Latency: Why Milliseconds Matter

In industrial environments, latency can mean the difference between a good product and a defective one, or between safe operation and an accident.

Latency Requirement Examples

Sources of Delay

• Local Network: 0.1-1 millisecond
• Edge Gateway: 1-10 milliseconds
• Internet: 20-100 milliseconds (depending on geographic location)
• Cloud Processing: 10-500 milliseconds (depending on complexity)

Practical Example: Hybrid Vibration Monitoring System

Let us build a hybrid system for monitoring bearing vibration in a factory:

Components

At the Edge:

• Vibration sensors (IFM VVB001) on 20 motors
• Industrial edge computer (Siemens IOT2050)
• Local database (SQLite or SurrealDB embedded)

In the Cloud:

• Time-series database (InfluxDB Cloud)
• Machine learning model for predictive maintenance
• Grafana dashboard

Data Flow

1. Every millisecond: The sensor reads vibration (2000 samples per second)
2. Every second: The edge device computes locally: RMS, peak, frequency spectrum (FFT)
3. Immediately: If RMS exceeds the threshold, a local alert is triggered
4. Every minute: A summary (RMS, peak, temperature) is sent to the cloud
5. Every hour: The machine learning model analyzes trends and predicts failures
6. Daily: A bearing health report is displayed on the dashboard

Results

• Instant Alerts: Within seconds of detecting abnormal vibration
• Predictive Maintenance: Predicting bearing failures 2-4 weeks in advance
• Reduced Downtime: 60% decrease in unplanned downtime
• Bandwidth Savings: Only 1% of raw data sent to the cloud

Summary

Edge computing and cloud computing are not alternatives but complements. The edge provides response speed and autonomy, while the cloud provides analytical power and storage. Hybrid architecture combines the best of both worlds: immediate decisions at the machine and advanced analytics in the cloud. Latency is the primary criterion for choosing where to process. Start by identifying the requirements of each application, then design an architecture that places processing in the most appropriate location.